Rafael Gomes

Led a production AI lead-gen pipeline that converts unstructured web data into structured, scored business intelligence, replacing 100% of manual workflows with continuous automation. Improved enrichment precision from 82% to 96% by combining deterministic pre-extraction with LLM review/correction, and implemented resilient multi-provider LLM abstractions using Zod-validated structured outputs.

Architected a stateful order-management and fulfillment controller using deterministic state machines across the full lifecycle with no manual intervention. Modeled the operational data layer in PostgreSQL (Drizzle + serverless Neon) and replaced polling with cryptographically verified, event-driven webhook flows to prevent forged-callback issues.

Built the full stack for a high-concurrency platform reaching 23,000+ active users and winning the Sonic Mobius Hackathon. Implemented WebSocket-based real-time distributed state sync with sub-second updates, optimized low-end mobile performance using a centralized 8Hz tick model, and improved incident response via Sentry observability across web and mobile.

Developed a real-time asset-tracking engine in NestJS/TypeScript with sub-second streaming latency, scaling to thousands of concurrent connections via Server-Sent Events. Reduced user-facing API latency by ~60% using Redis read-through caching and targeted PostgreSQL query optimization, and improved reliability with systematic unit testing.

Architected interconnected Node.js services processing 10k+ monthly transactions, including automated reconciliation and real-time analytics that replaced manual user calculations. Improved high-traffic dashboard load time by ~40% using virtual scrolling, code splitting, and query optimization, and unified multiple third-party data sources into a single platform view.

Built a task-management system with an automated social-verification pipeline that applies configurable filters to reject non-compliant actions instantly, removing manual moderation from the reward flow. Developed real-time engagement analytics dashboards to surface behavioral patterns from raw event logs.

Built a multi-module healthtech SaaS that replaced clinical records, scheduling, and invoicing with a single integrated platform used daily by clinic teams. Implemented a self-healing invoice automation queue with exponential backoff and distributed locks, and added domain-isolated encryption and row-level locking to eliminate double-confirmation bugs.

MBA in Software Engineering (in progress) at USP / ESALQ with expected completion in 2026.

Bachelor of Engineering (B.Eng.) in Electrical & Electronics Engineering from UFSC from 2014 to 2020.